Our method
We did not train a model. We built a library.
Most AI ad tools tell you their model was trained on millions of ads. Ours was not, and neither was theirs — that is not how these systems work. Here is what NEXA actually does, with the numbers you can hold us to.
1,362
Real ads in the reference library
36
Brands they came from
653
Of them video, analysed shot by shot
9
Sectors, each with its own playbook
These figures are read from the database when this page is built, not typed into it. They move when the library does.
Briefed against, not trained on
Every ad in the library is analysed, not stored as a mood board. We record how it is built — the angle it argues, the hook it opens on, where the subject sits, how it is lit, what the copy claims and how long it ran. An ad that has been live for months is telling you something an ad that ran for a week is not.
When you generate, a director reads your Brand Memory and searches that library. It looks at the closest references in your sector, decides which are worth borrowing from, and writes a shot brief. Then it renders several, inspects the finished ads, and drops any that could not ship.
Your work never enters the library. The references are ours. Your Brand Memory, your products and everything NEXA makes for you stay yours, and are never used to improve what anyone else gets.
What we will not claim
- That we trained a model on these ads. We did not fine-tune anything. They are retrieved and read at the moment you generate — which is why we can tell you the exact count.
- An average performance lift. We have no way to measure what our creatives did in your account, so any percentage we printed would be invented.
- A star rating or a customer quote we do not have. Our testimonials section renders nothing until there is something real to put in it.
- That every ad in this library is a winner. Most have been running ninety days or more, which is the best public signal there is. It is a signal, not a guarantee.